Abstract:

A multivariant optimization algorithm (MOA) is proposed. The proposed method makes full use of the multi-core processors and the large memory of modern computers. Multivariant searchers (atoms) explore the solution space and remember the historical information selectively. The MOA gets its name from the multivariant characters of multiple searchers. Atoms are divided into global atoms and local atoms according to variant responsibilities. Global atoms explore the whole solution space to discover potential areas. Local atoms exploit potential areas for a local refinement. Theoretically, the MOA is proved to be accessible to the global optimal solution. Experiments based on benchmark functions show that the MOA has competitive performance compared with other methods in terms of accessibility.